Capacity Metrics¶
The Capacity Metrics dashboard shows utilization, throttling and the interactive-versus-background CU mix of your Fabric and Premium capacities, from the Microsoft Fabric Capacity Metrics app.
Open it with Select Analysis > Capacity Metrics.
Before you begin¶
This page has data only when MetaMiner reads at least one Fabric Capacity Metrics semantic model:
- The Microsoft Fabric Capacity Metrics app is installed and refreshed in your tenant. See Capacity Metrics App.
- MetaMiner's service principal, or a security group it belongs to, is a Contributor or Admin on the workspace that hosts the app's semantic model.
- The semantic model ID is entered in the Power BI environment and the capacities to collect are ticked. See Capacity Metrics Datasets.
- The Fabric Capacity Metrics Data step of Validate the Connection passes.
Trial capacities (SKU starting with FT) are not collected unless pbi.fcm.trial.capacities.enabled=true is set in server\MetaMinerServer.properties.
TODO: list the supported Fabric Capacity Metrics app versions
MetaMiner reads two column layouts of the app's Capacities table. Confirm which app versions these correspond to and list them here.
How it works¶
The capacity collector queries the Capacity Metrics semantic model every hour and stores capacities, items, hourly metrics, 30-second CU detail and storage. On first collection it reaches back 14 days. The capacity rollup job summarises the data every 5 minutes. See Data Freshness.
This page does not use the common range and environment filters. It has its own:
| Control | Values | Default |
|---|---|---|
| Period | 1h, 6h, 24h, 7d, 14d, 30d | 24h |
| Capacities | (All), or individual capacities (hover for SKU and region) | (All) |
The selection is kept in the page address.
Image of the Capacity Metrics dashboard, period 7d: the Period and Capacities bar, three KPI cards and the Per-Capacity Scorecard with one row highlighted red
../images/dashboards/capacity-metrics.png
KPI cards¶
| Card | What it shows |
|---|---|
| Capacities | Capacities in the current selection |
| Avg Utilization | CU-weighted average utilization across the selection. Amber from 80 percent, red from 100 percent. |
| Throttled | Capacities with interactive delay above zero in the period |
Panels¶
| Panel | What it shows | Click |
|---|---|---|
| Per-Capacity Scorecard | Per capacity: Capacity, SKU, Util %, Peak %, Delay (s), IRej (s), BRej (s), P95 Delay, P95 IRej, P95 BRej, CU mix (int / bg). Rows in red had interactive-delay throttling in the period. | A row opens the spike drill-down |
| Utilization Ribbon | Hourly utilization per capacity. 100 percent is the capacity limit; 120 percent is where throttling starts. | - |
| Throttle-Seconds Trend | Interactive delay seconds per day, all selected capacities | - |
| CU Mix by Capacity | Interactive against background CU share per capacity (blue = interactive, orange underline = background) | - |
| Recurring Pressure Heatmap | Peak utilization by weekday x hour of day across the selected capacities. Darker cells mean higher recurring peaks. | - |
| Spike Drill-Down | For one capacity, the 30-second windows around its worst delay: Time (UTC), CUs, Util %, Delay %, IRej %, Pk6 Inter. Blue line = utilization, red = interactive delay. | - |
| Column | Meaning |
|---|---|
| Delay | Seconds of interactive delay (throttling that slows user queries) |
| IRej | Seconds of interactive rejection (user requests refused) |
| BRej | Seconds of background rejection (refreshes and other background jobs refused) |
| P95 ... | 95th-percentile intensity of the same measure over the period |
"P95 columns are temporarily unavailable; summable metrics are unaffected." appears while percentile data is being prepared.
Image of the Spike Drill-Down for one capacity: the chart of 30-second windows with the utilization and delay lines, and the table below
../images/dashboards/capacity-spike.png
Questions this page answers¶
- Which capacities run hot, and how often do they throttle users or refreshes?
- When does pressure recur (weekday and hour), so heavy refreshes can move?
- Is a capacity busy with interactive use or with background work?
- What exactly happened during a throttling spike?
Settings¶
| Property | Default | Effect |
|---|---|---|
web.apps.pbiusage.fcm.enable |
true |
false turns the Capacity Metrics page off; its panels then show no data |
web.apps.pbiusage.fcm.stale.hours |
3 |
Hours after which the capacity data counts as stale |
web.apps.pbiusage.fcm.max.hours |
2160 |
Longest period, in hours (90 days), a capacity request may cover |
pbi.fcm.trial.capacities.enabled (MetaMinerServer.properties) |
false |
Also collect trial capacities |
Troubleshooting¶
No capacity CU data in this window
No Capacity Metrics data was collected for the selected capacities and period. Check the prerequisites above and the capacity metrics items in Collection Problems.
A capacity is missing from the list
It is not ticked in the environment's capacity selection, it is a trial capacity, or it no longer exists in the Capacity Metrics app (MetaMiner then stops collecting it).